Transcript Real time parking availability estimation

Real time street parking availability
estimation
Dr. Xu, Prof. Wolfson, Prof. Yang, Stenneth, Prof. Yu
University of Illinois, Chicago
• In one business district, vehicles searching for
parking produces 730 tons of CO2, 47000 gallons
on gasoline, and 38 trips around the world.
2
Problem
• estimating street parking availability using
only mobile phones
• mobile phone distribution among drivers
• GPS errors, transportation mode detection errors,
Bluetooth errors, etc.
3
Motivations
•
•
•
•
save time and gas to find parking
reduce congestion and pollution
mobile phone are ubiquitous
affordable - SF park 8000 parking spaces cost
23M USD
• external sensors such as cameras not utilized
4
Why mobile phones ?
• ubiquitous with several sensors (GPS, gyro,
accelerometer)
• several people own a mobile phone
• other alternatives
– Sensor in pavement (e.g. SF Park) $300 + $12 per
month
– Manual reporting (e.g. Google OpenSpot)
– Ultrasonic sensors on taxi (e.g. ParkNet) $400 per
sensor
5
Contributions
• parking status detection (PSD)
• street parking estimation algorithms
– historical availability profile construction (HAP)
– parking availability estimation (PAE)
•
•
•
•
weighted average (WA)
Kalman Filter (KF)
historical statistics (HS)
scaled PhonePark (SPP)
6
PSD, HAP, PAE
7
Parking status detection (PSD)
• Determine when/where a driver park/deparks
Image sources: http://videos.nj.com/, http://pocketnow.com/smartphone-news/
http://sf.streetsblog.org
8
Parking Status Detection (PSD)
• We proposed three schemes for PSD
– transportation mode transition of driver
– Bluetooth pairing of phone and car
– Pay by phone piggyback
9
3 Schemes for PSD
Transportation mode transition
(GPS/accelerometer)
Bluetooth
Pay-by-phone piggy
back
10
HAP construction
• estimates the historic mean (i.e.𝑞(t)) and
variance (i.e. 𝑄(t)) of parking
• relevant terms
– prohibited period, permitted period
– false positives, false negatives
– b, N
11
Why is Building Profile Non-trivial
• Low sample rate due to low market
penetration
– 1% to 5%
• Errors in parking status detection
– False negative
• Missing parking activities that have occurred
• E.g., misclassifying parking as getting off a bus
– False positive:
• Reporting parking activities that have not occurred
• E.g., misclassify getting on a bus as deparking
Historical availability profile (HAP) Algorithm
• Start with a time at which the street block is fully available,
e.g., end of a prohibited time interval (start permitted period)
• When a parking report is received, availability is reduced by:
b: penetration ratio
(uniform distribution)
1  fp
fp: false positive probability
b  (1  fn )
fn: false negative probability
Justification:
1. Each report (statistically) corresponds to 1/b actual parking
2. 1/(1fn) reports should have been received if there were no false negatives
3. The report is correct with 1fp probability
• Similarly when a deparking report is received
HAP algorithm
PP1
PP2
PPm
m

qˆ ( t ) 
m
aˆ i ( t )
i 1
m
Qˆ ( t ) 

2
( aˆ i ( t )  qˆ ( t ))
i 1
m
14
HAP uncertainty bounding
• Given an error tolerance, with what P the diff
between q(t) and 𝑞(t) is less than x parking
spaces.
• Lemma 1
• Lemma 2
15
More specifically:
Cumulative distribution function of normal distr.
Prob {| qˆ ( t )  q ( t ) |  }  2   ( 
Estimation average
True average
m
) 1
Qˆ ( t )
Estimation variance
• Example:
– If we want error < 2 with 90% confidence,
• standard deviation of the estimation is 10 (i.e., the average
fluctuation of estimated availability at the 8:00am is 10).
– then we need 68 permitted periods.
• i.e. about two months of data.
Number of
samples , or
permitted
periods
Parking Availability Estimation (PAE)
• Solely real time observations
– scaled PhonePark (SPP) – capped
• Solely historical parking data (HAP)
– historical statistics (HS)
𝑥(t) = 𝑞(t)
17
Parking Availability Estimation (PAE)
• Combining history with real time
– Weighted average
𝑥(t) = 𝑤𝐻𝑆 𝑞(t) + 1 − 𝑤𝐻𝑆 𝑎(t)
RMSE of estimated mean
1
0.9
b=1%, fn=fp=0,
Chestnut
0.8
b=1%, fn=fp=0.1,
Chestnut
0.7
b=50%, fn=fp=0, Polk
0.6
b=50%, fn=fp=0.1, Polk
0.5
0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
b=50%, fn=fp=0.25,
Polk
wHS
18
Parking Availability Estimation (PAE)
• combining history with real time
– Kalman Filter estimation (KF)
𝑥(t) =
𝑅 𝑡
.𝑞(t)
𝑅 𝑡 +𝑄(𝑡)
𝑄 𝑡
+
.𝑎(t)
𝑅 𝑡 +𝑄(𝑡)
19
Evaluation
• RT data from SFPark.org 04/10 to 08/11
• Polk St (12 spaces )and Chestnut St (4 spaces )
20
HAP Results
• RMSE between q(t) 𝑎𝑛𝑑 𝑞(t)
• b = 1% , see for b = 50% in paper
Polk St. block
12 spaces available
Chestnut St. block
4 spaces available
21
PAE results
• RMSE between x(t) 𝑎𝑛𝑑 𝑥(t)
• b =1 % , see for b = 50% in paper
2.5
0.54
2
WA
1.5
KF
SPP
1
HS
0.5
RMSE of estimated availability
RMSE of estimated availability
0.53
0.52
0.51
WA
0.5
0.49
KF
0.48
SPP
0.47
HS
0.46
0.45
0
0.44
fn=fp=0.05
fn=fp=0.15
fn=fp=0.25
fn=fp=0.05
fn=fp=0.15
fn=fp=0.25
22
PAE results
• Boolean availability i.e. at least one slot available
• b =1 %
0.8
0.95
0.85
0.8
WA
0.75
KF
0.7
SPP
0.65
HS
0.6
boolean availability accuracy
boolean availability accuracy
0.9
0.75
0.7
WA
0.65
KF
SPP
0.6
HS
0.55
0.55
0.5
0.5
fn=fp=0.05
fn=fp=0.15
fn=fp=0.25
fn=fp=0.05
fn=fp=0.15
fn=fp=0.25
23
Related work
• ParkNet
• SFPark.org project
$400 per system
for each vehicle
$300 per sensor +
$12 per month
service. Project cost
$23 million
• Google’s OpenSpot
Cumbersome
Image sources: http://www.thesavvyboomer.com/
http://pocketnow.com/smartphone-news/
http://sf.streetsblog.org
27
Conclusion
• schemes for parking status detection (PSD)
– GPS, accelerometer, Bluetooth
• historical availability profile (HAP) algorithm
• real time parking availability estimation
algorithms (PAE)
25
Acknowledgements
• SF Park team (J. Primus etc.)
• Reviewers for fruitful comments
• NSF and NURAIL
26